Time-Reversion Fast-Sampling Score-Based Model for Limited-Angle CT Reconstruction.

Area of Science:

• Medical Imaging
• Computational Imaging
• Artificial Intelligence

Background:

• Score-based generative models (SGMs) are effective for limited-angle computed tomography (LACT) reconstruction.
• Traditional SGMs require extensive sampling steps, leading to high computational costs.
• Achieving rapid, high-quality LACT reconstruction with preserved details remains a challenge.

Purpose of the Study:

• To develop an innovative rapid-sampling strategy for SGMs in LACT reconstruction.
• To reduce the computational burden of SGM-based LACT while maintaining image fidelity.
• To introduce the time-reversion fast-sampling (TIFA) score-based model.

Main Methods:

• Proposed the time-reversion fast-sampling (TIFA) model grounded in robust optimization theory.
• Implemented a rapid-sampling mechanism including jump sampling, time-reversion with re-sampling, and compressed sampling.
• Utilized numerical simulations, physical phantoms, and clinical cardiac datasets for evaluation.

Main Results:

• TIFA (200 steps) outperformed state-of-the-art methods (2000 steps) in LACT reconstruction for [0°, 90°] and [0°, 60°] datasets.
• TIFA successfully reconstructed high-quality images with as few as 10 sampling steps.
• Demonstrated superior performance in preserving sharp edges and small features.

Conclusions:

• The TIFA model offers a significant advancement in rapid and efficient LACT reconstruction.
• TIFA substantially reduces computational requirements without compromising image quality.
• This method holds promise for accelerating medical image analysis and diagnosis.